Smart AI Learns New Tricks Without Costly Redoes

MIT researchers have come up with a clever new way to teach artificial intelligence — or 'AI' — new things. Think of it like this: imagine you've trained a brilliant apprentice who knows how to do 90% of your job. Traditionally, if you wanted them to learn a completely new skill, you'd almost have to send them back to school for a full, expensive retraining.

This new method, called 'MeMo', means AI can pick up new knowledge and skills on the fly, without needing to be completely re-taught from scratch. It's a bit like giving your apprentice a small, targeted course on a new software program, rather than making them repeat their entire degree. For businesses that use AI, this could mean significant savings in time and money, as they won't need to completely rebuild or retrain their AI systems every time they need to adapt to a new task or piece of information.

The ability for AI to adjust and learn quickly is a big deal. For example, if you run a small business using an AI customer service chatbot, and you introduce a new product line, previous AI models might need extensive re-engineering to understand and answer questions about it. With MeMo, the AI could be updated much more efficiently, learning about the new products without forgetting everything else it knows. This makes AI much more adaptable and useful for ever-changing business needs.

While the original article talks about crypto projects, the underlying technology has broader implications. Any business looking to use AI for things like analysing data, assisting customers, or automating tasks could benefit from this. It means AI tools will be able to stay current and relevant for longer, without incurring huge costs to keep them updated. It’s about getting more out of your existing AI investments, making them more flexible and future-proof.